Nearly 7 in 10 workers say generative AI will help them better serve customers, so it’s no wonder companies are clamoring to understand how they can use the technology to their advantage.
Silvio Savarese, Executive Vice President and Chief Scientist for Salesforce AI, leads an AI Research team that aims to push the boundaries of generative AI and understand what the technology can do – without compromising trust.
Here, Savarese offers a preview of what his team is developing.
Q. How would you describe your team’s mission?
We pioneer cutting-edge, simple, and intuitive generative AI technology that drives Salesforce product innovation, empowers our customers, influences the global AI technology landscape, and positively transforms society.
Our AI research doesn’t live in a research and development vacuum. Our team serves as a force multiplier, solving critical pain points that support Salesforce’s business needs. We accomplish this by regularly collaborating with our product teams and customers to understand their requirements.
Our team serves as a force multiplier, solving critical pain points that support Salesforce’s business needs.
This collaboration has helped us define our strategic focus and establish a set of trust principles to ensure our generative AI solutions deliver accuracy, reliability, confidentiality, privacy, and security.
Q. Diving deeper, how are your generative AI projects influenced by Salesforce’s trusted AI principles?
Salesforce’s AI Trust Layer serves as the basis for our work to operationalize these principles. For example, we manage generative AI accuracy concerns by incorporating guardrails like grounding in customer data to help us prevent “hallucinations” by using trusted data sources. We also prioritize safety and honesty to ensure that AI-generated output remains unbiased and inoffensive.
Confidentiality and privacy are also a focal point, so we employ strict measures to block third parties from accessing protected personal data without our customers’ and users’ consent. And strong security protocols shield our AI models from possible cyberattacks, which further safeguard user information.
Q. How does generative AI address major business challenges such as service reliability?
Enterprises increasingly require on-demand availability, highly solid uptime, and reliable security. For example, when companies offer a service, customers expect it to be available most of the time, which powers trust.
My team builds AI tools to ensure Salesforce’s infrastructure and operations are always available, fault-tolerant, secure, and sustainable — tying back to our company’s number one value, which is trust. These tools proactively analyze complex processes to predict and manage incidents, learn the root causes of incidents, and identify mitigation strategies.
My team builds AI tools to ensure Salesforce’s infrastructure and operations are always available, fault-tolerant, secure, and sustainable — tying back to our company’s number one value, which is trust.
Q: How will generative AI help users digest information?
The sheer amount of ever-increasing content drives information overload, which presents a significant challenge for users.
In response, our team develops generative AI solutions that help users easily, efficiently, and intelligently optimize their data consumption. From enhancing their search results to effectively summarizing findings to helping them quickly answer questions, our tools eliminate the need to browse countless documents, conversations online, or from data repositories.
Q: How does sustainability play a role in the AI future, especially considering the potential compute resources needed to run some of these popular models?
A recent report showed that 75% of developers want to reduce the environmental impact of their work, which is especially relevant considering generative AI requires significant compute resources. In fact, training generative AI models can result in 600 tons of CO2 emissions.
Salesforce generative AI solutions incorporate the company’s core value of sustainability and enable users to achieve these goals. In fact, research found that Salesforce’s CodeGen model can increase performance by 2.5x, resulting in significantly more carbon-efficient code.
Q. What do you consider the “next step” in generative AI’s development?
Our team is at the forefront of defining the future of generative AI, evolving the technology from being a passive entity to an active player.
Our team is at the forefront of defining the future of generative AI, evolving the technology from being a passive entity to an active player.
Generative AI has traditionally been reactive, whereby the software takes action in response to a user’s query or request to produce something. The next generation of generative assistants will focus on proactive large action models (LAMs) that detect and solve problems autonomously — all on their own, without prompts. Generative AI will tackle tasks including testing and debugging code, collecting data by conducting internet searches, and much more for Service Cloud and Sales Cloud use cases.
These LAMs are poised to unleash a sea change in generative AI by automating and coordinating processes for businesses while also using their language skills to directly communicate with stakeholders and incorporate human feedback on the fly.
Thinking bigger, LAMs could even collaborate with other LAMs, with each focused on achieving a specific goal. This automated team could be led by a LAM “director,” who would coordinate their efforts and serve as their liaison to a human that has project oversight.
This thrilling frontier is generative AI’s next evolution, and I expect it will deliver incredible impact for businesses and industry-wide transformation.
Learn more
- Read this piece, penned by Savarese, about building generative AI we can trust
- Check out this blog to explore other key AI projects at Salesforce
- Find more information about generative AI at Salesforce
